Predicting malaria prevalence using climate data with artificial neural networks and ridge regression model

Abstract

This research proposes the implementation of an Artificial Neural Network (ANN) model to assist in predicting the incidence of malaria using climate variables as key predictors. The prevalence of malaria in Katsina State, Nigeria has been modelled using climate parameters that include temperature, humidity, rainfall and wind speed. The model was trained and validated using historical climate and malaria data spanning several years 2018–2023. The findings show that malaria transmission patterns could not be efficiently predicted via climatic variations, leading to poor performance of the model with a coefficient of determination (R²) calculated to be 0.9863 during training and −0.6551 during testing using ANN. Therefore, there is a need for a precise projection of malaria prevalence which is critical for health care planning and early preventative measures. The fact that transmission of malaria is impacted by seasonal cycles, variability in the climate, and delaying effects makes forecasting challenging using ANN. We therefore suggest a versatile statistical method that integrates seasonal harmonics, climate covariates, and autoregressive dynamics into a ridge-regularized regression. Residual bootstrap is used to quantify predictive uncertainty, and post-hoc calibration is used to address systematic bias. Our model shows good in-sample performance (RMSE = 1.600 × 10⁵, R² = 0.8764) and high correlation with out-of-sample observations (RMSE = 2.123 × 10⁵, R² = 0.8236), using 72 months of training data (2018–2022) and a 12-month test period (2024). Empirical coverage of 91.7% is attained by bootstrap-based 95% prediction intervals. Predictive accuracy is further enhanced by calibration and residual AR(1) adjustment. The proposed method is appropriate for accurate malaria forecasting, comprehensible, and computationally effective. These results show how early detection systems for malaria prevention and control can be supported by machine learning-based climate-health modelling.

1. Introduction

The goal of this research is to examine the complex relationship between malaria prevalences and climate change in Katsina State, Nigeria. It seeks to do this so as to provide information that will improve disease control strategies and to throw insight for the management of malaria. Through the analysis of climate data and mathematical modelling, this research contributes to a more robust, adaptable, and resilient strategy for predicting malaria under the influence of climate change. Therefore, we first used multi-year meteorological data and Artificial Neural Networks (ANNs) to build a data-driven technique for malaria prediction. The objective was to determine the extent to which climatic factors could explain and forecast malaria occurrences in the state. We also present a ridge-regularized seasonal regression model incorporating climate covariates, autoregressive components, and harmonic seasonality. Systematic bias is addressed by post-hoc calibration, and prediction uncertainty is measured using residual bootstrap. The model is evaluated for training and validation using monthly data on malaria incidence from 2016 to 2023 and 2024, respectively.

This page shows an excerpt of the introduction. The full text, including the theorems, proofs and examples, is available in the PDF.

About this article

Cite this article

Sanusi, U., Abba, S., Idris, I.M. & Yantumaki, A.L. (2026). Predicting malaria prevalence using climate data with artificial neural networks and ridge regression model. Asia Mathematika, 10(2), 28–41. https://doi.org/10.5281/zenodo.23086571

Publication history